My article titled, “Will artificial intelligence take your job or make you rich?” was published on Asia’s largest business newspaper- The Economic Times on 21st December, 2015 over here & on e27, one of Asia’s largest startup and technology blogs , over here.
Artificial intelligence: The past, present & the future
While it is already beyond doubt that machines can perform most tasks that require physical labour more quickly and efficiently than humans, it is becoming increasingly clear that computers are able to perform several tasks that require, for the lack of a better word, “intelligence” better than humans.
Tasks that require a narrow range of expertise, that can be boiled down to a linear progression of steps were mastered by computers ages ago e.g. calculating the 25th root of a 25 digit number. Over time, the machines got better at tasks that required more than just basic logic. In 1997 itself, an IBM supercomputer – Deep Blue beat the world’s best human chess player. Now, chess is a game that requires a whole lot more than just basic computation. It requires complex strategic thinking and decision making, even creativity – something that was thought to be the fore of humans. Deep Blue is an example of Artificial Narrow Intelligence i.e. a system that performs exceedingly well in a limited domain but is incapable of doing much in a different domain.
Over time, the machines have gotten better at more complex tasks such as data mining, financial modeling, drug discovery, DNA sequencing and even driving cars! Not only have they gotten better over time, they have been able to outperform humans at many of these tasks. They will replace humans at most of these tasks in the near future. In fact, a studies estimate that 47% of jobs in USA and 34% of the jobs in UK are at risk of being lost forever to machines.
While this may be terrible news to people whose jobs are at risk, it is great news for the shareholders of companies which have good A.I systems since they will be able to cut costs and drastically improve productivity & therefore profitability. Let us take the example of 1 industry and see how artificial intelligence (intelligent systems) is changing it forever – the banking industry.
The problem with Indian banks
“Banking is necessary, banks are not” – Bill Gates
10.7% of the loan-book of by the Indian banking system consists of stressed loans (GNPAs + restructured loans). This number has been varying roughly between 8% and 11% over the recent years indicating that there is something wrong with the way loans are being made. Slightly older data shows that SME loans and individual loans were stressed (GNPA + Restructured advances) at ~ 10.6% and ~8.8% respectively, completely demolishing the widespread notion that it was the “fat-cat” promoters of bankrupt or near-bankrupt companies that were responsible for most of the stressed assets of Indian banks. The size of the market is staggering. Just the credit gap (excess demand over supply) for MSME loans in India is estimated to be $260 billion p.a. while the demand for individual loans in India (personal loans, gold loans, car loans etc.) is estimated to be $120 billion p.a. So, clearly there is a problem which, if solved even partially, can yield rich dividends.
The problem can be broken up as follows:
- The Data – Traditional data that encompasses credit history is surprisingly hard to come by especially in a place like India. According to the World Bank, 73% of the world’s population and 78% of India’s population are not covered by any credit bureaus. Even when data is available, it is not always of high quality and is often incomplete. So, on what basis can credit decisions be made when relevant data is missing or incomplete?
- The Decision making process – The decision making process used in traditional lending is deeply flawed. For starters, it relies on a rudimentary and semi-linear credit score that captures only a component of the credit risk associated with an individual. Next, the decisions are made by humans who, by nature, are prone to cognitive biases and there making errors. Standard operating procedures help to some extent but also bring a new set of problems including taking away discretion from decision makers.
Can Artificial Intelligence (Intelligent systems) solve this problem?
Banks and financial institutions are slowly being challenged by intelligent systems in developed markets e.g. Kreditech & Kabbage which use such systems to make lending decisions based on a variety of data points ranging from the social media footprint of a borrower to his credit history. As of now, they tend to target individuals or businesses with limited credit history or sub-optimal credit scores. Conventional lenders simply cannot accurately assess the creditworthiness of these borrowers and therefore either do not lend to them or lend at exorbitant rates sometimes as ridiculous as 100% p.a. (APR).
These intelligent systems use a wide variety of machine learning algorithms and apply them on a diverse set of new-age data points. Studies have shown that a person’s social media footprint, call records and even online shopping habits are good predictors of his creditworthiness (when viewed together). Therefore, they can be used to make credit decisions in the absence of conventional credit histories or even to augment the existing process.
Many of these intelligent systems find patterns hidden in the vast amounts of data available to them. Seemingly innocuous metrics such as the amount of time a person’s Twitter account has been active for, the time he spends reading the “Terms and Conditions” section of an online form are the number of telephone calls he makes per day are all, when viewed together & in totality, robust indicators of creditworthiness.
The best part is that this approach is working! These intelligent systems have default rates that tend to be lower than the industry averages – a fact that is reflected in the rapid growth of these companies’ loan portfolios, some even in the order of half a billion dollars. There are over 10 different companies in Europe and USA that use intelligent systems to make credit decisions and there is no doubt that more will come up over time and perhaps in a decade or so start giving banks a serious run for their money.
The road ahead
FinTech in India has largely been about payments thus far but it is a matter of time before lenders that use intelligent systems to make credit decisions become mainstream players. There are several startups working in this space even in India, with my own Monsoon FinTech being one of them. New age data is one part of the solution but even without it, machine learning based intelligent systems have been demonstrated to often make better decisions than the current SOP-based model. The simplest way for an NBFC or a bank to verify this statement is to get a robust intelligent system to learn on half of their existing loan book and verify its performance on paper (profitability) on the other half.
Banks and NBFCs in India are trying to embrace new technologies with the largest bank – SBI being surprisingly proactive. The results thus far are very encouraging. India poses unique challenges but there are a surprisingly large number of low-hanging fruits that are ripe for the picking. Even incremental value added could fetch hefty payoffs with the math being heavily loaded in the favor of new technologies. I will leave the reader with a simple math problem. If an NBFC or a bank has an impaired assets ratio of 8% of its loan-book and an intelligent system manages to perform 20% better than its current system (an incremental improvement), what would be the boost to the profitability of the loan-book? Therein lies the answer to the question – Will artificial intelligence take your job or make you rich? (Hint- The answer is not 20 %.)
(Ashwini Anand, CFA is a Co-Founder of Monsoon Fintech, an Indian startup that builds & leverages on intelligent systems to help lenders make profitable loans. He used to work with investment banks such as Merrill Lynch, Bank or America and Barclays Capital before this. He can be contacted on LinkedIn here )